For the National Institutes of Health-AARP (NIH-AARP) Diet and Health Study: Cancer incidence data from the Atlanta metropolitan area were collected by the Georgia Center for Cancer Statistics, Department of Epidemiology, Rollins School of Public Health, Emory University. Cancer incidence data from California were collected by the California Department of Health Services, Cancer Surveillance Section. Cancer incidence data from the Detroit metropolitan area were collected by the Michigan Cancer Surveillance Program, Community Health Administration, State of Michigan. The Florida cancer incidence data used in this report were collected by the Florida Cancer Data System under contract to the Department of Health (the views expressed herein are solely those of the authors and do not necessarily reflect those of the contractor or Department of Health). Cancer incidence data from Louisiana were collected by the Louisiana Tumor Registry, Louisiana State University Medical Center in New Orleans. Cancer incidence data from New Jersey were collected by the New Jersey State Cancer Registry, Cancer Epidemiology Services, New Jersey State Department of Health and Senior Services. Cancer incidence data from North Carolina were collected by the North Carolina Central Cancer Registry. Cancer incidence data from Pennsylvania were supplied by the Division of Health Statistics and Research, Pennsylvania Department of Health (the Pennsylvania Department of Health specifically disclaims responsibility for any analyses, interpretations, or conclusions). Cancer incidence data from Arizona were collected by the Arizona Cancer Registry, Division of Public Health Services, Arizona Department of Health Services. Cancer incidence data from Texas were collected by the Texas Cancer Registry, Cancer Epidemiology and Surveillance Branch, Texas Department of State Health Services. Cancer incidence data from Nevada were collected by the Nevada Central Cancer Registry, Center for Health Data and Research, Bureau of Health Planning and Statistics, State Health Division, State of Nevada Department of Health and Human Services.
For the Breast Cancer Detection Demonstration Project Follow-up Study: We are grateful to the study participants. We thank Dave Campbell and Leslie Carroll at Information Management Services for data management support.
For the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial: We thank Drs. Christine Berg and Philip Prorok, Division of Cancer Prevention, National Cancer Institute; the Screening Center investigators and staff of the PLCO Cancer Screening Trial; Mr. Tom Riley and staff at Information Management Services, Inc.; and Ms. Barbara O'Brien and staff at Westat, Inc. Most important, we acknowledge the study participants for their contributions to making this study possible.
For the US Radiologic Technologists (USRT) Study: We are grateful to the radiologic technologists who participated in the USRT Study; Jerry Reid of the American Registry of Radiologic Technologists for continued support of the study; Diane Kampa, Allison Iwan, and Richard Hoffbeck of the University of Minnesota for study coordination, data collection, and data management; and Jeremy Miller of Information Management Services for biomedical computing support.
Nulliparity is associated with an increased risk of endometrial cancer. It is less clear whether nulliparity modifies the association between other established hormone-related risk factors. The proportion of nulliparous women has increased since the mid-1970s, but most individual studies to date have been too small to test the hypothesis that endometrial cancer risk factors may be associated more strongly with risk among nulliparous women compared with parous women.
Data were aggregated on 26,936 postmenopausal, Caucasian, nulliparous women (360 endometrial cancers) and 146,583 postmenopausal, Caucasian, parous women (1378 endometrial cancers) from 4 US prospective studies (1979-2006). Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated in stratified analyses.
The risk of endometrial cancer was higher among nulliparous women than among parous women, as expected (nulliparous vs parous: HR, 1.42; 95% CI, 1.26-1.60). Stratified associations between endometrial cancer and hormone-related risk factors did not differ between nulliparous versus parous women: For both groups, oral contraceptives and earlier menopause were associated with reduced risk. The highest HRs were for obesity: A body mass index ≥30 kg/m2 (vs <25 kg/m2) increased the risk of endometrial cancer 3-fold among nulliparous women (HR, 3.04; 95% CI, 2.34-3.94) and parous women (HR, 2.88; 95% CI, 2.52-3.29).
Parous women are 20% to 40% less likely than nulliparous women to develop endometrial cancer.1-3 The exact mechanism by which parity reduces risk is not known, but several hypotheses have been proposed. Elevated progesterone levels during pregnancy may inhibit estrogen-driven endometrial cell proliferation and promote the differentiation and apoptosis of endometrial cells.4-6 Vaginal delivery itself or the postpartum involution of the uterus may facilitate the shedding of precancerous or cancerous cells in the endometrial lining of the uterus.7, 8 Certain infertility conditions, such as anovulatory disorders, that lead to nulliparity also may contribute to higher endometrial cancer risks among nulliparous women.9
These mechanisms, which are hypothesized to account for the protective effects of parity among parous women, occur during the reproductive years, but most endometrial cancers are diagnosed many years later, after menopause.10 Therefore, parity may shift the trajectory of endometrial cancer risk by modifying how other risk factors, such as oral contraceptive (OC) use, age at menopause, obesity, and menopausal hormone therapy (MHT), exert their risk-increasing or risk-decreasing effects on the uterus. We hypothesized that hormone-related endometrial cancer risk factors may be associated more strongly with risk among nulliparous women, who are not exposed to the protective effects of a full-term pregnancy. Although the prevalence of nulliparity among US women in their 40s has doubled in the last 30 years—from 10% in the mid-1970s to 18% in 200811—most individual studies have insufficient statistical power to formally evaluate potential risk-factor heterogeneity by parity. Therefore, we conducted a large aggregated analysis of data from 4 prospective studies to quantify and compare the associations between established hormone-related endometrial cancer risk factors among postmenopausal nulliparous women and parous women.
MATERIALS AND METHODS
This aggregated analysis included 4 National Cancer Institute (NCI) prospective studies: the Breast Cancer Detection Demonstration Project (BCDDP) Follow-Up Study12; the National Institutes of Health-AARP (NIH-AARP) Diet and Health Study13; the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial14; and the US Radiologic Technologists (USRT) Study.15 Each study used self-administered questionnaires or telephone-based interviews to collect information about hormone, reproductive, and other exposures, as described previously.12-15 Each study has obtained informed consent from participants and has been approved annually by the Special Studies Institutional Review Board of the NCI.
Within each study, we restricted the analysis to postmenopausal Caucasian women because of limited numbers of non-Caucasians in each of the 4 studies. Women who reported 1 or more live births or provided an age at first birth were classified as parous. Women who reported no live births and reported no age at first birth were classified as nulliparous. Women who underwent a hysterectomy before baseline, had unknown hysterectomy status at baseline, or had a personal history of cancer (other than nonmelanoma skin cancer) were excluded. We included the approximately 11% of the BCDDP cohort participants whose personal history of cancer was unknown at baseline.
Ascertainment and Classification of Endometrial Cancer
The 4 cohorts used self-report, retrieval of medical and pathology reports, and linkage with state cancer registries and the National Death Index to identify cancers that were diagnosed during follow-up.12-15 A summary of the study-specific methods was provided in a previous article in which we aggregated these 4 studies.16 From each study, we identified incident first primary endometrial cancers based on International Classification of Diseases ninth revision (ICD-9) codes 182.0 and 182.9 and ICD-10 codes C54.0, C54.1, C54.2, C54.3, C54.8, C54.9, or C55.9. Uterine corpus cancers with a histologic classification of 880017 (uterine sarcomas) or higher were excluded.
We estimated hazard ratios (HRs) and 95% confidence intervals (CIs) by using multivariable Cox regression methods (Proc PHREG; SAS version 9.1; SAS Institute Inc., Cary, NC). Age was the underlying time metric. In the NIH-AARP and PLCO studies, in which all women were aged ≥50 years at baseline and most were postmenopausal, follow-up for the current analysis began at the age of women on the baseline questionnaires. In the BCDDP and USRT cohorts, which included younger and premenopausal women, follow-up began at the first questionnaire (baseline or follow-up) on which a participant reported that she was postmenopausal. Person-time accrued until the diagnosis of endometrial or any cancer (excluding nonmelanoma skin cancer), death, loss to follow-up, age 85 years, or the study end date (ie, administrative censoring),12-15 whichever occurred first. Analyses in the BCDDP study incorporated self-reported hysterectomy during follow-up as an additional censoring criterion.
After Wald chi-square tests of study-specific HRs for hormone-related risk factors among parous women revealed no evidence of heterogeneity across studies, we aggregated data from all studies into a single data set. Then, we fit separate models to nulliparous women and parous women to obtain parity-stratified HRs for hormone-related risk factors. From each stratum-specific model, we obtained log-HRs and the covariance matrix of the estimates for each level of a given hormone-related factor (eg, categories of age at menopause). Heterogeneity between estimates for nulliparous and parous women was assessed using a Wald-type chi-square test based on properly chosen statistical contrasts between the parameter vectors of stratum-specific log-HR estimates standardized by their variance-covariances. Models assessing body mass index (BMI) were stratified by MHT, and vice versa, based on previously reported interactions between these exposures.18-21 To estimate the joint associations between nulliparity and BMI, OCs, and MHT, we fit models that included all women (nulliparous and parous) and used parous women in the lowest category of exposure (eg parous women with BMI <25 kg/m2) as the reference group.
Fully adjusted models included ages at menarche (ages <13 years, ≥13 years) and menopause (ages <45 years, 45-49 years, 50-54 years, ≥55 years), baseline BMI according to measured or self-reported height and weight (<25 kg/m2, 25 to <30 kg/m2, ≥30 kg/m2), ever use of OCs (never, ever), cumulative duration of MHT use at baseline (never, <5 years, ≥5 years), diabetes (no, yes), smoking (never, former, current), study, and calendar years of birth (<1933, ≥1933) and of study entry (<1995, ≥1995). Models with parous women also included the number of births. All covariates were modeled categorically with a separate category for missing data. The hazard plots for all of these covariates revealed no violations of proportional hazards assumptions. Outlier values for age at menopause (<30 years or >64 years) and BMI (<16 kg/m2) were considered missing. All significance tests were 2-sided, with P values <.05 considered statistically significant.
The 4 individual cohorts (Figure 1) collected baseline and follow-up data during an era of substantial changes in how MHT was prescribed and used. By 1996,22 it became clear that unopposed estrogen should be used only by women who had undergone hysterectomy. Because information about hormone formulation was not available from all baseline questionnaires, we used calendar year to classify baseline MHT use. Because all women in the current analysis had an intact uterus at the start of follow-up (see above), we assumed that all current MHT use reported in 1996 or after was estrogen plus progestin (E + P). To minimize the potential influence of the markedly elevated risks of endometrial cancer that persist after cessation of unopposed estrogen use,12 we limited the analysis of MHT to include only 1) NIH-AARP participants who reported never-use or current MHT use at baseline (1995-1996) and 2) PLCO and USRT participants who entered those studies on or after January 1, 1996 and reported never or current use. In addition, analyses of risk associated with E + P were administratively censored at June 30, 2002, because none of the cohorts had enough subsequent follow-up data to characterize the unprecedented and widespread decline in overall MHT use after July 2002.23
Table 1. Study Characteristics and the Risk of Endometrial Cancer Among Postmenopausal Caucasian Nulliparous Women Compared With Parous Women
No. Women (No. of Cases)
Mean Age at Baseline, y
HR [95% CI]: Nulliparous vs Parous
Abbreviations: BCDDP, the Breast Cancer Detection Demonstration Project Follow-Up Study; NIH-AARP, the National Institutes of Health-AARP Diet and Health Study; PLCO, the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial; USRT, the US Radiologic Technologist Study.
Study-specific estimates were adjusted for birth year and entry year (BCDDP, PLCO, and USRT), age at last menstrual period, age at menarche, body mass index, oral contraceptive use, menopausal hormone therapy use, diabetes, and smoking status.
Aggregated estimates are adjusted for birth year, calendar year of entry, study, age at menopause, age at menarche, body mass index, oral contraceptive use, menopausal hormone therapy use, diabetes, and smoking status.
The pooled data set included 360 incident endometrial cancers among 26,936 nulliparous women and 1378 cancers among 146,583 parous women (Table 1). Histologic type was available for approximately 50% of cancers, and most were endometrioid adenocarcinomas. The aggregate HR for nulliparity was 1.42 (95% CI, 1.26-1.60), and study-specific HRs ranged between 1.31 and 1.67 (Table 1).
There was no evidence of statistically significant heterogeneity by parity for any of the hormone-related risk factors (Figs. 2, 3). Endometrial cancer rates were approximately 25% lower among women who reported a history of OC use (Fig. 3a). Endometrial cancer risks increased with older age at menopause, increasing BMI, and longer duration E + P use. The largest HRs, approximately 3-fold, were observed for BMI ≥30 kg/m2 compared with <25 kg/m2. For the risk factors illustrated in Figures 2 and 3, similar patterns were observed among parous women who had 1 or 2 births and those who had ≥3 births (data not shown).
The association between BMI and endometrial cancer was strongest among women who reported no history of E + P use at baseline, as expected (Table 2). In this group, obesity was associated with a more than 5-fold increased risk of endometrial cancer. Similarly, the association between E + P use and endometrial cancer was strongest among the leanest women (data not shown). Consistent with the overall analyses, analyses stratified for BMI and E + P use revealed no significant heterogeneity between nulliparous women and parous women (all P values for heterogeneity were ≥ .05).
Table 2. Hazard Ratios and 95% Confidence Intervals for the Association Between Body Mass Index and Endometrial Cancer Risk in the Aggregated Sample of Postmenopausal Caucasian Women Stratified by Parity and Menopausal Hormone Therapy Use: Subset of Women Who Reported Never or Current Menopausal Hormone Therapy Use at Entry Into the National Institutes of Health-AARP Diet and Health Study and women who entered the PLCO Cancer Screening Trial or the US Radiologic Technologists Study after 1995 and reported never or current use at entry.
Separate models were fit to nulliparous and parous women. The models were adjusted for age at menarche, age at menopause, and oral contraceptive use as well as birth year, study, calendar year of entry, diabetes, and smoking status.
HRs also were adjusted for the number of births.
P values for the heterogeneity between parous and nulliparous women were based on the Wald-type chi-square test.
Among never users of menopausal hormone therapy
25 to <30
Among current users of menopausal hormone therapy
25 to <30
Models that included all women jointly quantified the associations between nulliparity and 3 modifiable risk factors: BMI, OC use, and cumulative duration of E + P use at baseline (Fig. 4a-c). Compared with parous women who had a BMI <25 kg/m2, nulliparous women who were obese at baseline had a 4-fold increased risk of endometrial cancer (HR, 4.03; 95% CI, 3.31-4.91) (Fig. 4a). Nulliparous women who had used OCs had a small, nonsignificant risk of endometrial cancer compared with parous women who had never used OCs (HR, 1.12; 95% CI, 0.88-1.43) (Fig. 4b). The joint association between long-term E + P use at baseline and nulliparity was 2.81 (95% CI, 2.13-3.72) (Fig. 4c).
The associations illustrated in Figures 2 and 3 were robust in sensitivity analyses, including the use of calendar time as the underlying time metric, complete case analyses, and censoring follow-up at age 80 years. Information about hysterectomy during follow-up was available only from the BCDDP. In this cohort, censoring women at hysterectomy (“censoring”) or ignoring the hysterectomy (“delayed censoring”) did not markedly alter the results presented above.
Our pooled analysis of 360 cases of endometrial cancer among 26,936 nulliparous women and 1378 cases among 146,583 parous women indicates that associations between endometrial cancer and hormone-related risk factors do not differ between nulliparous and parous women. We observed no statistically significant interactions between parity and age at menarche, OC use, BMI, age at menopause, or MHT use; and there were no differences in associations among parous women who had 1 or 2 births versus 3 or more births.
We had hypothesized that endometrial cancer risk factors may be more strongly associated with risk among women who did not experience the protective effects of a full-term pregnancy. In the breast, it is believed that the unique hormonal milieu of pregnancy induces persistent structural changes, such as the differentiation of terminal duct units, leaving the breast tissue less susceptible to carcinogenesis and rendering parous women at reduced risk of developing breast cancer.24 Because parous women are similarly less likely than nulliparous women to develop endometrial cancer, an analogous process plausibly may occur in the uterus. In addition, any parity-induced molecular or structural changes that affect long-term susceptibility to cancer may be even stronger in the uterus than in the breast, because the uterus may experience even more substantial changes than the breast during pregnancy, delivery, and postpartum. Hypothesized explanations for the protective effect of parity on endometrial cancer include higher levels of progesterone during pregnancy,4 “clearance” of premalignant lesions or cancerous cells during childbirth,7 or postpartum involution of the uterus.8 Our analysis did not directly test those hypotheses, but our results suggest that the potential effects of other hormone-related risk factors on endometrial cancer risk are not modified by parity status.
Previous studies that reported risk-factor differences between nulliparous and parous women reported inconsistent findings,25-31 but most studies included fewer than 100 endometrial cancer cases among nulliparous women. The low prevalence of nulliparous women, particularly in older studies, makes this question well suited for data pooling. Our pooled data set included a larger sample size and broader and more detailed risk-factor data, which, together, reduce the chances of spurious false-negative or false-positive findings. Consistent with our results, some of the earlier and smaller studies reported no evidence of interaction between parity and BMI,19, 29 menopausal estrogen therapy use,26, 28 or OC use.27, 30, 31 Our results did not replicate a previously reported interaction, which was not statistically significant, between parity and OC use.25
In addition to our large sample size and detailed risk-factor data, our analysis was strengthened by including prospective studies, each of which used sound methods of data collection and follow-up. Important limitations also exist. Even with our large sample size, some exposure categories had small sample sizes. Because most of the participants in these cohorts were Caucasian, we restricted the analyses to this group, which potentially limits the generalizability of our findings to other racial and ethnic groups.32 We did not investigate whether the null results further differed among type I cancers versus type II cancers. Hormone-related exposures may be more important for type I tumors than type II tumors,33 but the differences between these 2 tumor groups do not indicate that hormone-related risk factors affect the risk of only type I tumors. We focused on multiplicative interactions between nulliparous and parous women, on the assumption that potential differences would be greatest between these groups. More subtle differences may exist according to the number of births among parous women, but our analyses revealed similar associations in analyses stratified by parity among parous women. Gravidity also may affect potential differences by parity, on the assumption that hormone effects that are present in a full-term pregnancy also may be present but attenuated among women with incomplete pregnancies. Testing that hypothesis would require detailed data on the duration of incomplete pregnancies. Precise data on the duration of incomplete pregnancies is often not collected in epidemiologic studies, and that level of data was not available in the cohorts that were included in our pooled study. The extent of any such further modification of risk by hormone or other exposures because of incomplete pregnancies was therefore unknown in our analysis. We also lacked sufficient data to further stratify analyses according to the reason for nulliparity, such as infertility. Future studies with detailed infertility data are needed to better understand whether or how biologic factors related to infertility affect the subsequent risk of endometrial cancer, particularly among subgroups, such as women who give birth after a previous period of infertility.
The changing patterns of MHT use during the years in which these 4 cohort studies enrolled participants and collected data made it challenging to reconcile and pool the divergent data collected in the 4 studies. We lacked sufficient detail to assess risks by specific formulation and regimen, which is particularly important for endometrial cancer. Endometrial cancer is strongly associated with unopposed estrogen, but increasing evidence points to moderately elevated risks among women who use E + P for long durations.12, 18, 20, 34-37 Our aggregated analysis used calendar year to make assumptions about whether self-reported current MHT use was unopposed estrogen or E + P. Future studies with more recently collected data could evaluate this issue further.
In conclusion, the results from this large pooled analysis of data from 4 large prospective studies suggest that nulliparity does not modify the risks associated with established hormone-related endometrial cancer risk factors. Continued innovative investigations of potential differences in risk factors across at-risk groups of interest may help to advance the understanding of how parity reduces the risk of endometrial cancer.
We are indebted to the participants in the NIH-AARP Diet and Health Study for their outstanding cooperation. We also thank Sigurd Hermansen and Kerry Grace Morrissey from Westat for study outcomes ascertainment and management and Leslie Carroll at Information Management Services for data support and analysis.
This research was supported by the Intramural Research Program of the National Institutes of Health and the National Cancer Institute. Dr. Greenlee was supported in part by N01-CN-25,518.
The Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial was supported by contracts from the Division of Cancer Prevention, National Cancer Institute, National Institutes of Health, Department of Health and Human Services.